Please see inline.. > [Tom P.] Well, this isn't a very interesting case -- the app has > voluntarily switched to a lower rate? Yes that's right -- I just gave an example on how a CCID-3 sender could grow its sending rate.. >The interesting case is when the > app has been forced to switch to the lower rate, because of negative > feedback, but now the negative feedback has stopped. How do you tell the > difference between just being well-balanced and new capacity is available? Well some thoughts on how to achieve this: 1) Use Quickstart to see if capacity available -- app could inform DCCP -- which in turn can trigger a QS request..If request is granted, then app switches to a higher rate. But you need a QS enabled path.. 2) Trial and error method: App could verify from the transport that its getting positive feedback -- and app decides to increase its media rate to the next granularity. If transport successfully delivers it -- app continues at that rate for some time before increasing to the next rate. If transport fails, then stick with the same old rate for some period of time..You may end up losing packets -- but I see it would be a RTT worth of packets?? 3) If recent path state cache information is available, you could use that to decide. 4) As Tom suggested, use layered coding. 5)Another controversial possible method is transport could satisfy the app by increasing the rate to the media rate (abruptly) (like jumpstart) and see if it gets negative feedback-- if it gets negative feedback -- then start behaving conservatively..Few RTTs of aggressiveness may be OK provided you behave conservatively in the presence of negative feedback. With the current discussions at ICCRG on removing the need for TCP-friendliness and moving control to the network etc, this may be OK?? > I agree (with Arjuna's statement below) that padding is not an appetizing > idea, but is there any other way to tell the difference between the two > cases without offering up some sacrificial data? I too agree. Any of the above methods cant guarantee that app information wont be lost in the presence of negative feedback. You cant have the cake and eat the cake you see ;). I would be interested to see if someone comes up with a proposal to do this without padding. Regards Arjuna
